Post on 04-Aug-2020
transcript
DCM for evoked responses
Ryszard Auksztulewicz
SPM for M/EEG course, 2019
Does network XYZ explain my data better than network XY?
Which XYZ connectivity structure best explains my data?
Are X & Y linked in a bottom-up, top-down or recurrent fashion?
Is my effect driven by extrinsic or intrinsic connections?
Which neural populations are affected by contextual factors?
Which connections determine observed frequency coupling?
How changing a connection/parameter would influence data?
input
context
Collect data
Build model(s)
Fit your model
parameters to
the data
Pick the best
model
Make an
inference
(conclusion)
The DCM analysis pathway
Collect data
Build model(s)
Fit your model
parameters to
the data
Pick the best
model
Make an
inference
(conclusion)
The DCM analysis pathway
Data for DCM for ERPs / ERFs
1. Downsample
2. Filter (e.g. 1-40Hz)
3. Epoch
4. Remove artefacts
5. Average
Per subject
Grand average
6. Plausible sources
Literature / a priori
Dipole fitting
Source reconstruction
Collect data
Build model(s)
Fit your model
parameters to
the data
Pick the best
model
Make an
inference
(conclusion)
The DCM analysis pathway
Collect data
Build model(s)
Fit your model
parameters to
the data
Pick the best
model
Make an
inference
(conclusion)
The DCM analysis pathway
‘Hardwired’
model features
Models
Kiebel et al., 2008
Neuronal (source) model
spm_fx_erp
Inhib
Inter
Spiny
Stell
Pyr
L2/3
L4
L5/6
NEURAL MASS MODEL
Canonical Microcircuit Model (‘CMC’)
Bastos et al. (2012) Pinotsis et al. (2012)
Original
microcircuit
Updated
microcircuit
Canonical microcircuit
(predictive coding)Reduced model
(DCM)
mv
xv
spm_fx_cmcspm_fx_erp
Inhib
Inter
Spiny
Stell
Pyr
L2/3
L4
L5/6
NEURAL MASS MODEL CANONICAL MICROCIRCUIT
Pyr
Spiny
Stell
Inhib
Inter
Pyr
Canonical Microcircuit Model (‘CMC’)
Supra-
granular
Layer
Infra-
granular
Layer
Granular
Layer
Granular
Layer
Supra-
granular
Layer
Infra-
granular
Layer
Pinotsis et al., 2012
Canonical Microcircuit Model (‘CMC’)
Superficial Pyramidal
Cells
Spiny Stellate Cells
Deep Pyramidal
Cells
Inhibitory Interneurons
Granular
Layer
Supra-
granular
Layer
Infra-
granular
Layer
Pinotsis et al., 2012
Canonical Microcircuit Model (‘CMC’)
Superficial Pyramidal
Cells
Spiny Stellate Cells
Deep Pyramidal
Cells
Inhibitory Interneurons
Granular
Layer
Supra-
granular
Layer
Infra-
granular
Layer
3
2
5
6
89
Pinotsis et al., 2012
Canonical Microcircuit Model (‘CMC’)
Superficial Pyramidal
Cells
Spiny Stellate Cells
Deep Pyramidal
Cells
Inhibitory Interneurons
Canonical Microcircuit Model (‘CMC’)
Granular
Layer
Supra-
granular
Layer
Infra-
granular
Layer
4
1
10
7
6
9
Pinotsis et al., 2012
3
2
5
8
Superficial Pyramidal
Cells
Spiny Stellate Cells
Deep Pyramidal
Cells
Inhibitory Interneurons
Canonical Microcircuit Model (‘CMC’)
Granular
Layer
Supra-
granular
Layer
Infra-
granular
Layer
4
1
)( 3pSAF
10
7
6
9
)( 7pSAB
Pinotsis et al., 2012
3
2
5
8
Superficial Pyramidal
Cells
Spiny Stellate Cells
Deep Pyramidal
Cells
Inhibitory Interneurons
Canonical Microcircuit Model (‘CMC’)
Granular
Layer
Supra-
granular
Layer
Infra-
granular
Layer
4
1
)( 3pSAF
10
7
6
9
)( 7pSAB
)( 3pSAF
)( 7pSAB
)( 3pSAF
)( 7pSAB
Pinotsis et al., 2012
3
2
5
8
Superficial Pyramidal
Cells
Spiny Stellate Cells
Deep Pyramidal
Cells
Inhibitory Interneurons
Canonical Microcircuit Model (‘CMC’)
Granular
Layer
Supra-
granular
Layer
Infra-
granular
Layer
4
1
)( 3pSAF
10
7
6
9
)( 7pSAB
)( 3pSAF
)( 7pSAB
)( 3pSAF
)( 7pSAB
U
Pinotsis et al., 2012
3
2
5
8
Superficial Pyramidal
Cells
Spiny Stellate Cells
Deep Pyramidal
Cells
Inhibitory Interneurons
Canonical Microcircuit Model (‘CMC’)
Granular
Layer
Supra-
granular
Layer
Infra-
granular
Layer
4
1
)( 3pSAF
10
7
6
9
)( 7pSAB
)( 3pSAF
)( 7pSAB
)( 3pSAF
)( 7pSAB
)( 7pS
U
Pinotsis et al., 2012
3
2
5
8
Superficial Pyramidal
Cells
Spiny Stellate Cells
Deep Pyramidal
Cells
Inhibitory Interneurons
2
4
7
4
8597102
4
48
87
2))()()((
pppSpSpSA
Hp
pp
F −−−−=
=
Voltage change rate: f(current)
Current change rate: f(voltage,current)
Pinotsis et al., 2012
Canonical Microcircuit Model (‘CMC’)
2
4
7
4
8597102
4
48
87
2))()()((
pppSpSpSA
Hp
pp
F −−−−=
=
David et al., 2006; Pinotsis et al., 2012
Voltage change rate: f(current)
Current change rate: f(voltage,current)
H, τ Kernels: pre-synaptic inputs -> post-synaptic membrane potentials
[ H: max PSP; τ: rate constant ]
S Sigmoid operator: PSP -> firing rate
Canonical Microcircuit Model (‘CMC’)
2
2
3
2
437187
2
24
43
2))()()(((
pppSpSpSA
Hp
pp
B −−−+−=
=
Canonical Microcircuit Model (‘CMC’)
Granular
Layer
Supra-
granular
Layer
Infra-
granular
Layer
4
1
)( 3pSAF
2
4
7
4
8597102
4
48
87
2))()()((
pppSpSpSA
Hp
pp
F −−−−=
=
2
3
5
3
65476157
3
36
65
2))()()()((
pppSpSpSpSA
Hp
pp
B −−−+−−=
=
2
1
1
1
23253113
1
12
21
2))()()()(((
ppCupSpSpSpSA
Hp
pp
F −−−−−=
=
10
7
6
9
)( 7pSAB
)( 3pSAF
)( 7pSAB
)( 3pSAF
)( 7pSAB
)( 7pS
U
3Lp y =
Pinotsis et al., 2012
3
2
5
8
Van Wijk et al., 2018
Collect data
Fit your model
parameters to
the data
Pick the best
model
Make an
inference
(conclusion)
The DCM analysis pathway
‘Hardwired’
model features
Build model(s)
2 1
4 3
5
2 1
4 3
5
Input
2 1
4 3
5
Input
2 1
4 3
5
Input
2 1
4 3
5
Input
2 1
4 3
5
Input
Factor 1
2 1
4 3
5
Input
Factor 1 Factor 2
Collect data
Build model(s)
Fit your model
parameters to
the data
Pick the best
model
Make an
inference
(conclusion)
The DCM analysis pathway
Fixed
parameters
Fitting DCMs to data
Fitting DCMs to data
50 100 150 200-1.5
-1
-0.5
0
0.5
1
1.5mode 1
50 100 150 200-1.5
-1
-0.5
0
0.5
1
1.5mode 2
50 100 150 200-1.5
-1
-0.5
0
0.5
1
1.5mode 3
50 100 150 200-1.5
-1
-0.5
0
0.5
1
1.5mode 4
50 100 150 200-1.5
-1
-0.5
0
0.5
1
1.5mode 5
50 100 150 200-1.5
-1
-0.5
0
0.5
1
1.5mode 6
50 100 150 200-1.5
-1
-0.5
0
0.5
1
1.5mode 7
50 100 150 200-1.5
-1
-0.5
0
0.5
1
1.5mode 8
time (ms)
trial 1 (predicted)
trial 1 (observed)
trial 2 (predicted)
trial 2 (observed)
0 50 100 150 200 250-0.01
-0.005
0
0.005
0.01
time (ms)
Observed (adjusted) 1
0 50 100 150 200 250-0.01
-0.005
0
0.005
0.01
channels
time
(ms)
Predicted
0 50 100 150 200 250-0.01
-0.005
0
0.005
0.01
time (ms)
Observed (adjusted) 2
0 50 100 150 200 250-0.01
-0.005
0
0.005
0.01
channels
time
(ms)
Predicted
H. Brown
Fitting DCMs to data
50 100 150 200-1.5
-1
-0.5
0
0.5
1mode 1
50 100 150 200-1.5
-1
-0.5
0
0.5
1mode 2
50 100 150 200-1.5
-1
-0.5
0
0.5
1mode 3
50 100 150 200-1.5
-1
-0.5
0
0.5
1mode 4
50 100 150 200-1.5
-1
-0.5
0
0.5
1mode 5
50 100 150 200-1.5
-1
-0.5
0
0.5
1mode 6
50 100 150 200-1.5
-1
-0.5
0
0.5
1mode 7
50 100 150 200-1.5
-1
-0.5
0
0.5
1mode 8
time (ms)
trial 1 (predicted)
trial 1 (observed)
trial 2 (predicted)
trial 2 (observed)
0 50 100 150 200 250-0.01
-0.005
0
0.005
0.01
time (ms)
Observed (adjusted) 1
0 50 100 150 200 250-0.01
-0.005
0
0.005
0.01
channels
time
(ms)
Predicted
0 50 100 150 200 250-0.01
-0.005
0
0.005
0.01
time (ms)
Observed (adjusted) 2
0 50 100 150 200 250-0.01
-0.005
0
0.005
0.01
channels
time
(ms)
Predicted
H. Brown
Fitting DCMs to data
1. Check your data
0 50 100 150 200-5
0
5x 10
-14
time (ms)
Observed response 1
channels
peri-
stim
ulus
tim
e (m
s)
Observed response 1
50 100 150 200 250
0
50
100
150
200
0 50 100 150 200-5
0
5x 10
-14
time (ms)
Observed response 2
channels
peri-
stim
ulus
tim
e (m
s)
Observed response 2
50 100 150 200 250
0
50
100
150
200
H. Brown
Fitting DCMs to data
1. Check your data
2. Check your sources
H. Brown
1. Check your data
2. Check your sources
3. Check your model
Model 1
V4
IPLA19
OFC
V4
IPLA19
OFC
V4
IPL
Model 2
V4
IPL
H. Brown
Fitting DCMs to data
Fitting DCMs to data
1. Check your data
2. Check your sources
3. Check your model
4. Re-run model fitting
H. Brown
Collect data
Build model(s)
Fit your model
parameters to
the data
Pick the best
model
Make an
inference
(conclusion)
The DCM analysis pathway
Friston et al., 2016
Collect data
Build model(s)
Fit your model
parameters to
the data
Pick the best
model
Make an
inference
(conclusion)
The DCM analysis pathway
Does network XYZ explain my data better than network XY?
Which XYZ connectivity structure best explains my data?
Are X & Y linked in a bottom-up, top-down or recurrent fashion?
Is my effect driven by extrinsic or intrinsic connections?
Which connections/populations are affected by contextual factors?
input
context
Garrido et al., 2008
Example #1: Architecture of MMN
Garrido et al., 2007
Example #2: Role of feedback connections
Boly et al., 2011
Example #3: Group differences
Auksztulewicz & Friston, 2015
Example #4: Factorial design & CMC
Bastos et al., Neuron 2012
Attentioncf. Feldman & Friston, 2010
L2/3
L4
L5/6
smx
m
xx
FORWARD PREDICTION ERROR
BACKWARD PREDICTIONS
A1ST
G
xx
p
2x2 design:
Attended vs unattended
Standard vs deviant
(Only trials with 2 tones)
N=20
Auksztulewicz & Friston, 2015
Flexible factorial design
Thresholded at p<.005 peak-level
Corrected at a cluster-level pFWE<.05
Contr
ast estim
ate
Attention Expectation
Auksztulewicz & Friston, 2015
Flexible factorial design
Thresholded at p<.005 peak-level
Corrected at a cluster-level pFWE<.05
Contr
ast estim
ate
A1E1 A1E0 A0E1 A0E0
Auksztulewicz & Friston, 2015
inputinput
inputinput
Inh
Int
input
SP
input
Connectivity structure
Extrinsic modulation
Intrinsic modulation
Auksztulewicz & Friston, 2015
Example #5: Same paradigm, different data
Phillips et al., 2016
Example #5: Same paradigm, different data
Phillips et al., 2016
Example #6: Hierarchical modelling
Rosch et al., 2017
Evoked response potentials at Fz
Example #6: Hierarchical modelling
Rosch et al., 2017
Useless
data
Perfect
model
Useless
results
Perfect
data
Useless
model
Useless
results
Motivate your assumptions!
Thank you!
Karl Friston
Gareth Barnes
Andre Bastos
Harriet Brown
Hayriye Cagnan
Jean Daunizeau
Marta Garrido
Stefan Kiebel
Vladimir Litvak
Rosalyn Moran
Will Penny
Dimitris Pinotsis
Richard Rosch
Bernadette van Wijk